Modeling of flip-chip interconnects using novel neural network approaches

In this paper, we present an efficient neural network approach to analyze the bump structure of flip-chip transitions. 3D FEM simulations are used to extract the network parameters such as s-parameters and then the equivalent lumped circuit model is used to extract L-C and R. Validation of the equivalent circuit model is conducted by the comparison between the circuit simulation and the full wave simulation. Using the extracted data, training sets of the neural networks are composed. The proposed neural network, in which all L-C and R parameters are used simultaneously at the one output layer, shows nice prediction performance for the s-parameter characteristics.